Study On Extraction Of Crop Planting Structure Based On Spectral And Texture Features

Crop planting structure is closely related to the management of water using in irrigation area.Crop planting structure refers to the types and spatial distribution information of crops in the region.The dynamic information of crop planting structure is obtained by interpreting satellite remote sensing images,and then the water using and water demand of crops are discriminated based on the data of rainfall and crop growth change,which can provide data support for irrigation water management.But at present,there are two problems in the extraction of crop planting structure: one is that a single recognition feature cannot meet the needs of crop planting structure.For example,low-resolution images need the comparison of multi-temporal to complete the extraction of crop planting structure,and the classification accuracy is low in images without optimal window period;while high-resolution images can obtain higher accuracy,but the timeliness is poor,so dynamic monitoring cannot be realized.The other is when multi-feature is used to extract crop planting structure,the multi-feature lacks of effective feature screening which leads to too many classification indicators,finally it will affects the rate of classification.In order to solve the problems in crop extraction process,this paper based on the WFV image of the domestic high-score satellite,mainly includes four aspects:(1)Calculating the spectral and texture characteristics of different crops in the high-score WFV data,and analyzing the differences between the two features in the expression of different types of crops,so as to verify the importance in the classification process.(2)The CART algorithm is used to construct the random forest classification model for extracting crop planting structure,and the optimal parameters are set to avoid the phenomenon of over-fitting or insufficient training,which affects the classification accuracy.(3)Based on the correlation and segregation of the indexes calculated by the Bhattacharyya distance,and sorted according to the results of calculation,the indexes with the greatest degree of discrimination are finally screened to form a set of optimum indexes to ensure that the classification rate is improved on the basis of the original accuracy.(4)Take Shijin Irrigation District of Hebei Province as an example to verify the feasibility and accuracy of the extraction method of planting structure.The main conclusions are as follows:(1)By comparing the differences in the values of spectral and texture features of different crops,the importance of the two types of features in extracting crop planting structure was verified.(2)Calculating the separability of the indicators based on the Bhattacharyya distance algorithm,sorting them according to the order from big to small and determining the number of indicators based on the method of experimental comparison.Finally,the optimal indicators set of crop classification is constructed to reduce the workload of classification and reduce the classification rate.(3)The extraction model of crop planting structure based on CART algorithm is constructed,and the optimal marginal benefit points of tree nodes are determined by experiments.Achieving high precision automatic extraction and greatly reducing human intervention.(4)Taking Shijin Irrigation District as an example,the advantages of the crop classification method proposed in this paper are fully verified.The highest classification accuracy and Kappa coefficient were 88.5115 and 0.8331 respectively.At the same time,compared with the classification of mult-feature combinations,the optimized time is shortened by 3 minutes and 06 s,and compared with the classification results of 2/8m image data of GF-1,it is verified that the optimized crop extraction effect can reach 2/8m classification effect,thus making up for the deficiency of 2/8m in time phase.